Autonomous AI Agents & Harnesses Leaderboard
Evaluating compound multi-step coding agents, CLI harnesses, and tool orchestration frameworks across real-world GitHub issue resolution, trajectory steps, and cost-per-task.
Compound AI Systems & Autonomous Agent Leaderboard
Multi-turn tool orchestration, SWE-bench Verified issue resolution, Open Agent Leaderboard generality, trajectory steps, and cost-per-resolved-task ($/task)
Agent Systems Leaderboard: Scaffolding, Tool Use & Unit Economics
Evaluating raw models tests knowledge retrieval and token prediction. In contrast, Agent Benchmarks evaluate complete autonomous systems (Antigravity, Claude Code, Cursor, Windsurf, OpenHands) across tool calling precision (MCP, bash, browser), multi-subagent orchestration, unit economics ($/task), and long-horizon task completion.
| # | Agent System & Architecture | Tools & Capabilities | Foundation Model | Benchmark Suite | Pass RatePass % | Cost / TaskCost | Trajectory StepsSteps | Tool PrecisionPrecision | Harness & ProvenanceHarness |
|---|---|---|---|---|---|---|---|---|---|
| No matching agent evaluations found for the selected filters. | |||||||||
About Đại Việt (daiviet.dev) — Decentralized AI Intelligence
Đại Việt is an open-source AI benchmark aggregator and real-time GitHub trending intelligence engine. Operating on automated 15-minute ingestion cycles, it unifies empirical frontier model evaluations, compound agent harnesses, token pricing telemetry, and breakout open-source code repositories.
15-Min Star Velocity
Tracks viral open-source repositories across GitHub with rolling star snapshots at 15-minute, 24-hour, and 7-day windows to detect breakthrough AI tools before they peak.
10+ Benchmark Suites
Aggregates SWE-bench Verified, LiveBench 2.0, Berkeley Function Calling (BFCL), LMSYS Arena human Elo, Stanford HELM, MTEB, Epoch AI, MATH-500, and IFEval with 95% confidence intervals and VRAM specs.
Agent Leaderboards
Evaluates compound agent systems (Claude Code, OpenHands, SWE-agent, Codex CLI) across real-world issue resolution, multi-step tool calls, and cost-per-resolved-task.
830+ Model Catalog
Normalized specifications and cost calculators covering input, output, and cache read pricing across frontier models from OpenAI, Anthropic, Google, and open-source models.